Perplexity
•
Sonar Pro
•
Released 
January 2025

Perplexity
Sonar Pro

Perplexity's higher-tier Sonar model for live web answers with citations and the family's largest 200K context window.

Modality:
Text
model ID
perplexity/sonar-pro

Output Speed *

0.00
tok/s

Intelligence Index *

0
/ 100

Context Window *

200000
tokens

Input price

18
Anytoken

Output price

90
Anytoken
Sonar Pro: Citation-Backed Web Answers with a 200K Context Window Sonar Pro is Perplexity's higher-tier search-augmented model, which searches the live web for each query and attaches citations and source URLs. It lists a 200K context window, against 128K for base Sonar, and returns about twice as many search results. It is built for complex, multi-part questions rather than long-form generation. The best fit is an answer engine, research assistant, or fact-sensitive Q&A product where current, verifiable information matters more than reasoning depth. Add Sonar Pro to your research stack

Performance

Grounded Answering Is the Strength, Not General Reasoning

Sonar Pro's strength is retrieval-backed answering. It searches per query, attaches citations, and offers 200K context and about twice the search results of base Sonar. Perplexity's launch post reports a SimpleQA F-score of 0.858 for Sonar Pro versus 0.773 for Sonar. That is a vendor-reported, short-form factuality result. It supports the retrieval-quality claim but says little about reasoning or long-form synthesis. In production, use it where freshness and source traceability matter, and test it on your own domain questions before relying on the headline score.

Benchmarks

Sonar Pro on Artificial Analysis: Fast and Concise, Mid-Tier Intelligence

<cite index="1-11,1-12">Artificial Analysis scores Sonar Pro 8 on its current Intelligence Index, above the median of 7 for comparable models.</cite> <cite index="1-4,1-5">It marks the model deprecated and only keeps benchmarking a default 10k-token workload.</cite> <cite index="5-4,5-12">An earlier snapshot measured about 123 output tokens per second and 1.2M tokens generated during evaluation, against a 3.9M average.</cite> That points to fast, concise output. Index versions differ, so the scores of 8 and 15 are not comparable. These tests do not measure search groundedness.

Output Speed

*
0.00
tok/s

Intelligence Index

*
0
/ 100

MMLU *

Broad world knowledge and problem-solving
0
%

GPQA *

PhD-level scientific reasoning across physics, biology, chemistry.
0
%

HLE *

Adherence to multi-step structured instructions.
0
%

LiveCodeBench *

Tool-calling reliability in long agentic loops.
0
%

Technical Specifications

What the model supports

Sonar Pro is a proprietary, text-output model with a 200K-token context window and an output cap of roughly 8K tokens. These are separate limits, and the output cap matters more in production. <cite index="18-12">The search_context_size setting controls how much web evidence is retrieved, separately from the model's context window.</cite> Long reports, large code generations, or lengthy structured exports will hit the output ceiling long before the input window. Plan multi-step generation or hand the retrieved facts to a long-output model.
Verified Specifications — 
Sonar Pro
*
Input modalities
Text
output modalities
Text
Context window
200000
 tokens
Maximum output tokens
8000
Reasoning
No
Knowledge cutoff
January 2025
Pricing (standard)
18
 AnyTokens in
 / 
90
 AnyTokens out

Limitations & Trade-offs

Where Sonar Pro falls short

1
8K output ceiling. Sonar Pro accepts 200K tokens of context but is listed with a maximum output of about 8K tokens per response. That caps long reports, extensive document drafts, and large structured exports. Workloads that need long generations should use Sonar Pro only for retrieval and facts, then pass them to a model with a larger output limit.
2
Not a reasoning model. Artificial Analysis classes Sonar Pro as non-reasoning and rates its intelligence as mid-tier, and it points to Sonar Reasoning Pro as a newer option. Complex multi-step analysis, math, or planning over retrieved data will be better served by a reasoning model. Sonar Pro is strongest at retrieving and summarizing, not deliberating.
3
Retrieval-driven cost structure. <cite index="9-6">Sonar Pro carries a higher output token price and a request fee compared with other Sonar models.</cite> Every query triggers search, so cost scales with request count as well as tokens. High-volume, simple lookups are better handled by base Sonar, and tasks that do not need live data are better handled by a non-search model.
4
Benchmark and status caveats. The SimpleQA score is vendor-reported and measures short factual questions. <cite index="9-11">A 2026 Desearch benchmark scored sonar-pro below some specialised search providers on composite groundedness.</cite> Artificial Analysis also labels the model deprecated and suggests Sonar Reasoning Pro. Validate groundedness on your own queries and confirm current availability before committing.

Best-Fit Workloads

Where this model earns its place

01
Citation-backed answer engines Sonar Pro searches the web per query and returns citations and source URLs, with about twice the search results of base Sonar. That fits user-facing Q&A products where every answer must be traceable. The 8K output cap suits concise answers but not long articles, and cost scales with request volume.
02
Current-events and market research assistants Live web retrieval keeps answers current where a static model's training data would not. Sonar Pro suits analysts' questions on news, competitors, or regulations, and Pro-tier filters can narrow results by domain or timeframe. Treat outputs as sourced starting points and verify critical claims, since groundedness varies by benchmark.
03
Multi-turn research follow-ups The 200K context window is the largest in the Sonar family, so conversations can carry earlier turns and retrieved evidence further than on base Sonar's roughly 128K. This helps iterative research sessions with refinements. Total input and output must fit the window, and each reply is still limited to about 8K tokens.
04
Retrieval stage in a multi-model pipeline Sonar Pro can gather cited facts, which a reasoning or long-output model then turns into analysis or reports. This uses its strength, grounded retrieval, and avoids its limits in output length and reasoning depth. It suits research tools that separate fact gathering from synthesis.
Pricing in anytokens via AnyAPI
Input
18
₳
Output
90
₳
Cache write
—
₳
Cache read
—
₳

Integration

Access Sonar Pro via AnyAPI.ai

Access Sonar Pro through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate Sonar Pro without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access Sonar Pro and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test Sonar Pro against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use Sonar Pro from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use Sonar Pro for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

Sonar Pro has a 200,000-token context window, the largest in Perplexity's Sonar family. <cite index="14-7">OpenRouter lists its maximum output at 8,000 tokens.</cite> Context covers input and output combined, so long prompts leave less room, and each response is still capped at about 8K tokens.

Yes. <cite index="17-15">Sonar Pro searches the live web for each query and synthesizes an answer grounded in what it finds, with citations and source URLs.</cite> Perplexity describes it as returning about twice as many citations and search results as standard Sonar, which suits verifiable, source-attributed answers.

Sonar Pro is the higher tier. It has a 200K context window versus roughly 128K for Sonar, about twice the search results, and a higher Perplexity-reported SimpleQA F-score (0.858 versus 0.773). Sonar is lighter, cheaper, and quicker, so it fits simple, high-volume lookups better.

No. Artificial Analysis classes Sonar Pro as a non-reasoning model and scores it 8 on its Intelligence Index. Its strength is retrieval and cited summarization. For multi-step analysis, Perplexity offers Sonar Reasoning Pro, which adds chain-of-thought reasoning to search retrieval.

Sonar Pro pricing combines token charges with search or request fees, and output tokens cost more than on base Sonar. See the AnyTokens pricing component on this page for AnyAPI.ai rates. You can access Sonar Pro via API through AnyAPI.ai using the quickstart on this page.

* Benchmark data source: Artificial Analysis artificialanalysis.ai